Thanks.
I see conflicting info about whether noise was removed in GRD:
Yes: https://earth.esa.int/web/sentinel/technical-guides/sentinel-1-sar/products-algorithms/level-1-algorithms/ground-range-detected
Yes: Sentinel-1A GRD product
Yes No: Striping after terrain flattening
Strange is that, as I wrote in the original post, the output of noise removal is identical to the original GRD. I applied ‘Radiometric’->S1 Thermal Noise Removal’; then applied ‘Calibrate’ to both of original GRD and noise-sub GRD. The difference of the two is shown below. The flat grey is at grey level 122 everywhere, which corresponds to 0 dB difference…
Cut at a particular azimuth location along the range direction looks like:
